Yeah, those images explain it pretty well. There is one slight difference, though. Because the Hilbert/Z-order curves are defined recursively, algorithms for c

Yeah, those images explain it pretty well. There is one slight difference, though. Because the Hilbert/Z-order curves are defined recursively, algorithms for converting between the curve and row/column order are "cache-oblivious", in that they don't need to take an explicit block size. You write it once, and it performs well on any system regardless the CPU cache size and/or page size.